Google Duplex calls are sometimes made by real humans
While Google actually used the name Duplex for something different at first, it amazed the audience (and many across the world) when they debuted its re-branded service at Google I/O 2018. The feature uses A.I. and an, at times, eerily human-sounding computer voice to perform tasks including making reservations, setting appointments, and more. It’s taken a while for this feature to begin rolling out to the public but it’s here now and we have just learned that about 25% of the calls placed by Google Duplex are done by real humans.
It seemed that one of the things that surprised people the most about Google Duplex was its ability to sound so human. We’ve heard the typical computer voices before but Google has been working on making these sound incredibly fluid. Fluidity is only part of it though as the demonstration at Google I/O 2018 had the computer inserting “umms” and “uhhs” into the conversation just as a real human would. Since this feature has gone live, more and more companies are beginning to receive these phone calls.
Natt Garun of The Verge said he shadowed a number of restaurants in New York to see how they are reacting to this influx in A.I. phone calls. He said that most, if not all, of the people he spoke with, made sure to comment about how unmistakably human the computer sounded on the phone. This sounds normal at first as it was what so many people thought when they saw it in action at Google I/O. However, a new report from The New York Times revealed that about a quarter of all phone calls placed with Google Duplex are currently “started” by humans.
They also revealed that about 15% of Google Duplex phone calls start with an A.I. but then have a human intervene and take over due to various reasons. In fact, all Duplex calls are monitored by real humans. I can see how some would consider this to mean the feature has failed but A.I. needs to be trained in order to improve. It looks as if the Mountain View tech giant is handling this well instead of letting it run wild and ruin the end-user experience.